Upload
others
View
2
Download
0
Embed Size (px)
Citation preview
Electronic copy available at: http://ssrn.com/abstract=2020587Electronic copy available at: http://ssrn.com/abstract=2020587
1
Submitted to the EDI Quarterly
The Discovery of Price Responsiveness – A Survey of Experiments involving Dynamic Pricing of
Electricity
Ahmad Faruqui and Jenny Palmer1
Abstract
This paper surveys the results from 126 pricing experiments with dynamic pricing and time-of-use pricing
of electricity. These experiments have been carried out across three continents at various times during
the past decade. Data from 74 of these experiments are sufficiently complete to allow us to identify the
relationship between the strength of the peak to off-peak price ratio and the associated reduction in
peak demand or demand response. An “arc of price responsiveness” emerges from our analysis, showing
that the amount of demand response rises with the price ratio but at a decreasing rate. We also find
that about half of the variation in demand response can be explained by variations in the price ratio. This
is a remarkable result, since the experiments vary in many other respects – climate, time period, the
length of the peak period, the history of pricing innovation in each area, and the manner in which the
dynamic pricing designs were marketed to customers. We also find that enabling technologies such as
in-home displays, energy orbs and programmable and communicating thermostats boost the amount of
demand response. The results of the paper support the case for widespread rollout of dynamic pricing
and time-of-use pricing.
Introduction
Electric utilities, which run a capital-intensive business, could lower their costs of doing business by
improving their load factor. Other capital intensive industries, such as airlines, hotels, car rental
agencies, sporting arenas, movie theaters routinely practice a technique known as dynamic pricing to
improve load factor. In dynamic pricing, prices vary to reflect the changing balance of demand and
supply through the day, through the week and through the seasons of the year.
Congestion pricing, a simpler form of dynamic pricing, is used to regulate the flow of cars into central
cities. Parking spaces in most central cities are priced on a time-of-day basis and in some cities such as
San Francisco the prices are varying dynamically. In California, special lanes on freeways are priced
dynamically and the Bay Bridge charges toll on a time-of-use basis.
But it has been difficult for electric utilities to follow these examples. There has always been doubt that
electric users can change their usage patterns. To assuage these doubts, in the late 1970s and early
1980s, a dozen electricity pricing experiments were carried out with time-of-use rates in the United
1 The authors are economists with The Brattle Group, based in San Francisco. They are grateful to fellow
economist Sanem Sergici of Brattle for reading an early draft of this paper. Comments can be directed to [email protected].
Electronic copy available at: http://ssrn.com/abstract=2020587Electronic copy available at: http://ssrn.com/abstract=2020587
2
States. 2 They showed that customers do respond to such rates by lowering peak usage and/or shifting it
to less expensive off-peak periods. But smart meters that would charge on a time-of-day basis were
expensive in those days and little progress occurred in the ensuring years. Even now, less than one
percent of the more than 125 million electric customers in the United States are charged on a time-of-
use basis.
However, the California energy crisis of 2000-01 reinvigorated interested in dynamic pricing, not only in
that state but globally. Over the past decade, two dozen dynamic pricing studies featuring over one
hundred dynamic time-of-use and dynamic pricing designs were carried out across North America, in the
European Union and in Australia and New Zealand.3
These experiments have yielded a rich body of empirical evidence. We have compiled this into a
database, D-Rex, which stands for Dynamic Rate experiments. This contains the following data from
each pilot: details of the specific rate designs tested in the pilot, whether or not enabling technologies
were offered to customers in addition to the time-varying rates, and the amount of peak reduction that
was realized with each price-technology combination. The D-Rex results provide an important
perspective on the potential magnitude of impacts with different dynamic rate approaches and should
inform the public debate about the merits of smart meters and smart pricing. Across the 129 dynamic
pricing tests, peak reductions range from near zero values to near 60 percent values. However, it would
be misleading to conclude that there is no consistency in customer response.4
We focus on nine of the best designed, more recent experiments to examine the impact of the peak to-
off peak price ratio on the magnitude of the reduction in peak demand, or demand response. Because
the amount of demand response varies with the presence or absence of enabling technology, such as a
smart thermostat, an energy orb or an in-home display, we separate those pricing tests without and
with enabling technology. We find a statistically significant relationship between the price ratio and the
amount of peak reduction, and quantify this relationship with a logarithmic model. This relationship is
termed the Arc of Price Responsiveness. We find that for a given price ratio, experiments with enabling
technologies tend to produce larger peak reductions, and display a more price-responsive Arc.
Sidebar: The Dynamic Rates
2 For an early summary, see Ahmad Faruqui and J. Robert Malko, “The Residential Demand for Electricity by Time-
Of-Use: A Survey of Twelve Experiments with Peak Load Pricing,” Energy, Volume 8, Issue 10, October 1983. For more recent surveys, see Ahmad Faruqui and Jenny Palmer, “Dynamic Pricing and its Discontents,” Regulation, Fall 2011 and Ahmad Faruqui and Sanem Sergici, “Household Response to Dynamic Pricing of Electricity – A Survey of 15 Experiments,” Journal of Regulatory Economics, October 2010. Faruqui and Palmer also discuss the more common myths that surround legislative and regulatory conversations about dynamic pricing. 3 Most dynamic pricing studies have included multiple tests. For example, a pilot could test a TOU rate and a CPP
rate and it could test each rate with and without enabling technology. Thus, this pilot would include a total of four pricing tests. 4 See, for example, the concluding remarks in an otherwise excellent paper by Paul Joskow, “Creating a smarter
U.S. electrical grid,” Journal of Economic Perspectives, Winter 2012.
3
Time-of-Use (TOU). A TOU rate could either be a time-of-day rate, in which the day is divided into time
periods with varying rates, or a seasonal rate into which the year is divided into multiple seasons and
different rates provided for different seasons. In a time-of-day rate, a peak period might be defined as
the period from 12 pm to 6 pm on weekdays, with the remaining hours being off-peak. The price would
be higher during the peak period and lower during the off-peak, mirroring the variation in marginal costs
by pricing period.
Critical Peak Price (CPP). On a CPP rate, customers pay higher peak period prices during the few days a
year when wholesale prices are the highest (typically the top 10 to 15 days of the year which account for
10 to 20 percent of system peak load). This higher peak price reflects both energy and capacity costs
and, as a result of being spread over relatively few hours of the year, can be in excess of $1 per kWh. In
return, the customers pay a discounted off-peak price that more accurately reflects lower off-peak
energy supply costs for the duration of the season (or year). Customers are typically notified of an
upcoming “critical peak event” one day in advance but if enabling technology is provided, these rates
can also be activated on a day-of basis.
Peak Time Rebate (PTR). If a CPP tariff cannot be rolled out because of political or regulatory constraints,
some parties have suggested the deployment of peak-time rebate. Instead of charging a higher rate
during critical events, participants are paid for load reductions (estimated relative to a forecast of what
the customer otherwise would have consumed). If customers do not wish to participate, they simply
buy through at the existing rate. There is no rate discount during non-event hours. Thus far, PTR has
been offered through pilots, but default (opt-out) deployments have been approved for residential
customers in California, the District of Columbia and Maryland.
Real Time Pricing (RTP). Participants in RTP programs pay for energy at a rate that is linked to the hourly
market price for electricity. Depending on their size, participants are typically made aware of the hourly
prices on either a day-ahead or hour-ahead basis. Typically, only the largest customers —above one
megawatt of load — face hour-ahead prices. These programs post prices that most accurately reflect the
cost of producing electricity during each hour of the day, and thus provide the best price signals to
customers, giving them the incentive to reduce consumption at the most expensive times.
The Dynamic Pricing Studies
The D-Rex Database contains the results of 129 dynamic pricing tests from 24 pricing studies. 5 As shown
in Figure 1, these results range from close to zero to up to 58 percent. Part of the variation in impacts
comes simply from the fact that different rate types are being tested. Filtering by rate in Figure 2, some
trends begin to emerge. We observe that the Critical Peak Pricing (CPP) rate tends to have higher
impacts than Time-of-Use (TOU) rates, likely because the CPP rates have higher peak to off-peak price
ratios. We can also filter by the presence of enabling technology, as in Figure 3, and observe that for the
same rates, the impacts with enabling technologies tends to be higher.
5 23 of the 24 studies are pricing pilots. The other study is PG&E’s full scale rollout of TOU and SmartRate.
4
Figure 1. Impacts from Residential Dynamic Pricing Tests, Sorted from Lowest to Highest
Figure 2. Impacts from Pricing Tests, by Rate Type
Peak Reductions, Sorted from Lowest to Highest
Pricing Test
Pe
ak
Re
du
cti
on
1 129
0%
10%
20%
30%
40%
50%
60%
Peak Reductions by Rate
Pricing Test
Pe
ak
Re
du
cti
on
1 129
0%
10%
20%
30%
40%
50%
60%
TO
U
CP
P
PT
R
RT
P
VP
P
5
Figure 3. Impacts from Pricing Tests, by Rate Type and Presence of Enabling Technologies
Even with the rate and technology filters, there remains significant unexplained variation. In order to
understand the cause of this variation, we first limit the sample to only the best-designed studies which
have reported the relevant data. We selected studies in which samples are representative of the
population and the results are statistically valid. Moreover, we selected studies in which participants
were selected randomly, as opposed to volunteers responding to a mass mailing. The nine best-designed
pilots, shown in Table 1, include 42 price-only tests and 32 pricing tests with prices cum enabling
technology.6 In these 74 tests, the peak reductions range from 0% to just under 50%. The remainder of
this paper focuses on explaining the variation in these results.
6 OG&E was not included in these screened results because only the draft results are available thus far. When
these results are finalized, they will be included in this analysis.
Peak Reductions by Rate and Technology
Pricing Test
Pe
ak
Re
du
cti
on
1 129
0%
10%
20%
30%
40%
50%
60%
TO
U
TO
U w
/ T
ech
PT
R
PT
R w
/ T
ech
CP
P
CP
P w
/ T
ech
RT
PR
TP
w/ T
ech
VP
P w
/ T
ech
6
Table 1. Features of the Nine Dynamic Pilots
Utility Location Year Rates
Enabling Technologies
Number of Tests
Baltimore Gas & Electric Maryland
2008, 2009, 2010
CPP, PTR CPP w/ Tech, PTR w/ Tech
17
Connecticut Light & Power Connecticut 2009 TOU, CPP, PTR TOU w/ Tech, CPP w/ Tech, PTR w/ Tech
18
Consumers Energy Michigan 2010 CPP, PTR CPP w/ Tech 3
Pacific Gas & Electric (Full scale rollout)
California 2009, 2010 TOU, CPP Not tested 4
Pacific Gas & Electric, San Diego Gas & Electric, Southern California Edison (Statewide Pricing Pilot)
California 2003, 2004 TOU, CPP CPP w/ Tech 4
Pepco DC District of Columbia
2008, 2009 CPP, PTR, RTP2
CPP w/ Tech, PTR w/ Tech, RTP w/ Tech
4
Salt River Project Arizona 2008, 2009 TOU Not tested 2
Utilities in Ireland2 Ireland 2010 TOU TOU w/ Tech 16
Utilities in Ontario Ontario, Canada
2006 TOU, CPP, PTR Not tested 6
1. Run by the Commission for Energy Regulation (CER)
Total 74
2. The two RTP pricing tests are excluded from this analysis because they do not have a clear peak to off-peak price ratio.
7
Figure 4. Impacts from Pricing Tests, by Rate Type and Presence of Enabling Technologies
Methodology
The nine best-designed studies in D-Rex include 42 price-only test results and 32 price-cum-enabling
technology test results for a total of 74 observations. For each result, we plot the all-in peak to off-peak
price ratio against the corresponding peak reduction. As expected, the CPP and PTR rates tend to have
higher peak to off-peak ratios than the TOU rates, with some overlap, and those rates with higher price
ratios tend to yield greater peak reductions. 7 It also appears that that the enabling technology impacts
may be greater than those with price only.
7 For the PTR rate, the effective critical peak price is calculated by adding the peak time rebate to the rate that the
customer pays during that time period.
Peak Reductions of Screened Results
Pricing Test
Pe
ak
Re
du
cti
on
1 74
0%
10%
20%
30%
40%
50%
60%
8
Figure 5. Impacts from Pricing Tests by Peak to Off-Peak Ratio, Showing Rate Type and Presence of Enabling Technologies
The plot suggests that peak impacts increase with the price ratio but at a decreasing rate. The
logarithmic model would model rapid increases in peak reduction in the lower price ratios, followed by
slower growth.8
Logarithmic Model
( )
where y = peak reduction percent
Results
When we fit the logarithmic model to the full dataset (n = 74), it yields a coefficient of 0.106 with a
standard error of 0.012, significant at the 0.001 level. In other words, as the price ratio increases, the
peak reduction is also expected to increase. The peak-to-off-peak price ratio successfully explains 49
percent of the variation in demand response. The logarithmic curve suggests that if the peak to off-peak
price ratio were to get as high as 16, the peak reduction could be close to 30 percent.
8 We also considered a logistic growth model that features slow growth at lower price ratios followed by moderate
growth, followed by an upper bound peak reduction. The results were not significantly different with this functional form.
Peak to Off-Peak Price Ratio
Pe
ak
Re
du
cti
on
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
0%
10%
20%
30%
40%
50%
60%
PTR
PTR CPP
TOUTOU
PTR
CPPPTR
CPPCPP PTR
CPP
PTRTOU
CPP
TOU
CPP
TOU
CPP
PTR
TOU
TOU
TOUTOU
TOU
TOUTOU
TOU
TOU
TOU
TOU
TOU
PTRPTR
TOU
TOU
CPP
CPP
PTR
PTR
TOU
TOU
CPP
CPP
CPP
PTR
PTRPTR
PTR
PTR
PTR
TOUTOU
PTR
PTR
CPP
CPP
TOUTOU
PTR
PTR
CPP
CPP
CPP
CPP
PTR
TOUTOU
TOUTOU
PTR
PTR
PTR
PTR
With Enabling Technology
9
Figure 6. Impacts from Pricing Tests by Peak to Off-Peak Ratio with the Fitted Logarithmic Curve
We can narrow down the model to focus on the price-only observations separately from the enabling
technology observations. With this data, the model yields a coefficient of 0.077 with a standard error of
0.012, again significant at the 0.001 level. The coefficient is slightly lower here than in the full dataset,
suggesting that the impacts increase more slowly in the absence of enabling technology. In this case, the
adjusted R-squared value is 48 percent, meaning the ratio again explains almost half of the variation in
response. The logarithmic curve suggests that if the peak to off-peak price ratio were to get as high as
16, the peak reduction would be slightly over 20 percent.
With the enabling technology tests, we find that the curve has a steeper slope than the result with price-
only tests. The coefficient of the enabling technology curve is 0.130 which has a standard error of .02.
The regression successfully explains 53 percent of the variation in demand response. With a peak to off-
peak ratio of 16, the peak reduction is expected to be over 30 percent.
Full Dataset, n = 74
Peak to Off-Peak Price Ratio
Pe
ak
Re
du
cti
on
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
0%
10%
20%
30%
40%
Logarithmic Curve
10
Figure 7. Impacts from Pricing Tests by Peak to Off-Peak Ratio with the Fitted Logarithmic Curves, Segregated by Presence of
Enabling Technologies
The full regression results for the three different data specifications are shown in Table 2 below. In each
case, the coefficient on the natural log of the price ratio is positive and significant at the 0.001 level.
Table 2. Regression Results
Coefficient Full Dataset Price-Only
Enabling
Technology
Ln(Price Ratio) 0.10611 *** 0.07682 *** .13029 ***
(0.01254 )
(0.01220)
(0.02164 )
Intercept -0.01985
0.00654
-0.03668
(0.02234 )
(0.02071)
(0.04080 )
Adjusted R-Squared 0.4916
0.4852
0.532
F-Statistic 71.59
39.65
36.24
Observations 74
42
32
Standard errors are shown in parentheses below the estimates
*** = 0.001 significance
** = 0.01 significance
* = 0.05 significance
Price-Only (n = 42) and Enabling Technology (n = 32)
Peak to Off-Peak Price Ratio
Pe
ak
Re
du
cti
on
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
0%
10%
20%
30%
40%
Price-Only
Enabling Tech
Full Dataset
11
Conclusion
In our view, the results presented in this paper provide strong support for the deployment of dynamic
pricing. They conclusively show that customers are responsive to changes in the price of electricity. In
other words, they lower demand when prices are higher. Moreover, the results suggest that the
presence of enabling technology allows customers to increase their peak reduction even further. These
results may be used to quantify the potential peak reductions that may be expected when new dynamic
rates are rolled out and to monetize these benefits using estimates of the avoided capacity of capacity
and energy.9
Sources
Commission for Energy Regulation, “Electricity Smart Metering Customer Behaviour Trials (CBT) Findings
Report,” May 16, 2011.
eMeter Strategic Consulting, “PowerCentsDC Program: Final Report,” Prepared for the Smart Meter Pilot
Program, Inc., September 2010.
Faruqui, Ahmad and J. Robert Malko, “The Residential Demand for Electricity by Time-Of-Use: A Survey
of Twelve Experiments with Peak Load Pricing,” Energy, Volume 8, Issue 10, October 1983.
Faruqui,
Faruqui, Ahmad and Jenny Palmer, “Dynamic Pricing and its Discontents,” Regulation, Fall 2011.
Faruqui, Ahmad and Sanem Sergici, “BGE’s Smart Energy Pricing Pilot: Summer 2008 Impact Evaluation,”
Prepared for Baltimore Gas & Electric Company, April 28, 2009.
Faruqui, Ahmad and Sanem Sergici, “Impact Evaluation of BGE’s SEP 2009 Pilot: Residential Class
Persistence Analysis,” Presented to Baltimore Gas and Electric Company, October 23, 2009.
Faruqui, Ahmad and Sanem Sergici, “Impact Evaluation of NU’s Plan-It Wise Energy Program: Final
Results,” November 2, 2009.
Faruqui, Ahmad, Sanem Sergici and Lamine Akaba, “Impact Evaluation of the SEP 2010 Pilot,” Presented
to Baltimore Gas and Electric Company, March 22, 2011.
9 On the monetization of benefits arising from smart meters and dynamic pricing in the context of the
EU, see Ahmad Faruqui, Dan Harris, and Ryan Hledik, “Unlocking the €53 billion savings from smart
meters in the EU: How increasing the adoption of dynamic tariffs could make or break the EU’s smart
grid investment,” Energy Policy, 2010.
12
Faruqui, Ahmad, Sanem Sergici and Lamine Akaba, “Michigan Speaks—Consumer Energy’s Dynamic
Pricing Experiment,” CRRI Western Conference, Monterey California, June 16, 2011.
George, Stephen S., Josh Bode, Elizabeth Hartmann, “2010 Load Impact Evaluation of Pacific Gas and
Electric Company’s Time-Based Pricing Tariffs,” Prepared for Pacific Gas and Electric Company,
April 1, 2011.
George, Stephen S., Josh Bode, Mike Perry, Zach Mayer, “2009 Load Impact Evaluation for Pacific Gas
and Electric Company’s Residential SmartRate Peak Day Pricing and TOU Tariffs and SmartAC
Program,” Prepared for Pacific Gas and Electric Company, April 1, 2010.
IBM Global Business Services and eMeter Strategic Consulting, “Ontario Energy Board Smart Price Pilot
Final Report,” July 2007.
Kirkeide, Loren, “Effects of the EZ-3 (E-20) Time-of-Use Plan on Summer Energy Usage of Residential
Customers,” Salt River Project, unpublished paper, January 2012.
Biography of Authors
Ahmad Faruqui is a principal with The Brattle Group. He has been analyzing time-varying experiments
since the beginning of his career in 1979 and his early work is cited in the third edition of Professor
Bonbright’s canon on public utility ratemaking. The author of four books and more than a hundred
papers on energy policy, he holds a doctoral degree in economics from the University of California at
Davis and bachelor’s and master’s degrees from the University of Karachi.
Jennifer Palmer is a research analyst at The Brattle Group. Since joining The Brattle Group in 2009, she
has worked with a wide range of utilities on dynamic pricing and advanced metering projects. For
several utilities, she has developed dynamic tariffs, simulated the impacts of these rates on customer
consumption patterns, and estimated the resulting system-level benefits. She has a bachelor’s degree in
economics with a certificate in environmental studies from Princeton University.
13
Appendix
Impacts from Pricing Tests by Peak to Off-Peak Ratio, Showing Utility Names
Peak to Off-Peak Price Ratio
Pe
ak
Re
du
cti
on
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
0%
10%
20%
30%
40%
50%
60%
BGE
BGE BGE
CL&PCL&P
CL&P
CL&PCL&P
CL&PConsumers EnergyConsumers Energy
Pepco DC
Pepco DCPG&E
PG&E
PG&E
PG&E
SPP
SPP
BGE
Ireland
Ireland
IrelandIreland
Ireland
IrelandIreland
Ireland
Ireland
Ireland
Ireland
Ireland
BGEBGE
Ontario
Ontario
Ontario
Ontario
Ontario
Ontario
SRP
SRP
BGE
SPP
SPP
BGE
BGEBGE
BGE
BGE
BGE
CL&PCL&P
CL&P
CL&P
CL&P
CL&P
CL&PCL&P
CL&P
CL&P
CL&P
CL&P
Consumers Energy
Pepco DC
Pepco DC
IrelandIreland
IrelandIreland
BGE
BGE
BGE
BGE